Page 1
Medical Image Authentication through
Watermarking Preserving ROI
Dissertation
submitted in partial fulfillment of the requirements
for the degree of
Master of Technology, Computer Engineering
by
Sonika C. Rathi
Roll No: 121022005
under the guidance of
Prof. V. S. Inamdar
Department of Computer Engineering and Information Technology
College of Engineering, Pune
Pune - 411005.
June 2012
Page 2
Dedicated to
my mother
Smt. Kavita C. Rathi
and
my father
Shri. Chandrakant D. Rathi
Page 3
DEPARTMENT OF COMPUTER ENGINEERING AND
INFORMATION TECHNOLOGY,
COLLEGE OF ENGINEERING, PUNE
CERTIFICATE
This is to certify that the dissertation titled
Medical Image Authentication throughWatermarking Preserving ROI
has been successfully completed
By
Sonika C. Rathi
(121022005)
and is approved for the degree of
Master of Technology, Computer Engineering.
Prof. V. S. Inamdar, Dr. Jibi Abraham,
Guide, Head,
Department of Computer Engineering Department of Computer Engineering
and Information Technology, and Information Technology,
College of Engineering, Pune, College of Engineering, Pune,
Shivaji Nagar, Pune-411005. Shivaji Nagar, Pune-411005.
Date :
Page 4
Abstract
Telemedicine is a well-known application, where enormous amount of medical data
need to be securely transferred over the public network and manipulate effectively.
Medical image watermarking is an appropriate method used for enhancing security
and authentication of medical data, which is crucial and used for further diagnosis
and reference. This project focuses on the study of medical image watermarking
methods for protecting and authenticating medical data. Additionally, it covers
algorithm for application of water marking technique on Region of Non Interest
(RONI) of the medical image preserving Region of Interest (ROI).
The medical images can be transferred securely by embedding watermarks in
RONI allowing verification of the legitimate changes at the receiving end without
affecting ROI. Segmentation plays an important role in medical image processing
for separating the ROI from medical image. The proposed system separate the
ROI from medical image by GUI based approach, which works for all types of
medical images. The experimental results show the satisfactory performance of
the system to authenticate the medical images preserving ROI.
iii
Page 5
Acknowledgements
I express my sincere gratitude towards my guide Prof. V. S. Inamdar for her
constant help, encouragement and inspiration throughout the project work. Also
I would like to thank our Head of Department, Prof. Jibi Abraham for her able
guidance and for providing all the necessary facilities, which were indispensable
in the completion of this project.
I take this opportunity to express my hearty thanks to all those who helped me
in the completion of my project work. I am very grateful to the authors of various
articles on the Internet, for helping me become aware of the research currently
ongoing in this field.
I am very thankful to my parent for their constant support. I would also
like to thank Manisha Mantri, Dr. Dinesh Mantri and Pravin Reddy
for their valuable suggestions and helpful discussions. Last, but not the least, I
would like to thank my classmates for their valuable comments, suggestions and
unconditional support.
Sonika C. Rathi
College of Engineering, Pune
May 21, 2012
iv
Page 6
Contents
Abstract iii
Acknowledgements iv
List of Figures viii
1 Introduction: Digital Watermarking 1
1.1 Digital watermarking . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Principle of Digital Watermarking . . . . . . . . . . . . . . . . . . . 2
1.3 Types of Watermarking System . . . . . . . . . . . . . . . . . . . . 4
1.3.1 Visible watermarking system . . . . . . . . . . . . . . . . . . 4
1.3.2 Invisible watermarking system . . . . . . . . . . . . . . . . . 5
1.3.3 Blind watermarking system . . . . . . . . . . . . . . . . . . 5
1.3.4 Non-blind watermarking system . . . . . . . . . . . . . . . . 5
1.3.5 Robust watermarking system . . . . . . . . . . . . . . . . . 5
1.3.6 Fragile watermarking system . . . . . . . . . . . . . . . . . . 6
1.4 Properties of Digital Watermarking . . . . . . . . . . . . . . . . . . 6
1.4.1 Imperceptibility . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.2 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.3 Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.4 Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.5 Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.6 Invertibility . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.7 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Watermarking Techniques . . . . . . . . . . . . . . . . . . . . . . . 8
1.5.1 Spatial domain watermarking . . . . . . . . . . . . . . . . . 9
1.5.2 Frequency domain watermarking . . . . . . . . . . . . . . . 10
1.6 Application of Watermarking . . . . . . . . . . . . . . . . . . . . . 12
1.7 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Page 7
2 Medical Image Watermarking Introduction 14
2.1 Principle of Medical Image Watermarking . . . . . . . . . . . . . . 16
2.2 Requirements of Medical Image Watermarking . . . . . . . . . . . . 17
2.2.1 Imperceptibility . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3 Authenticity . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.4 Reversibility . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.5 Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.6 Intactness of ROI . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Literature Survey 19
3.1 Region of Interest (ROI) Segmentation . . . . . . . . . . . . . . . . 19
3.1.1 Magnetic Resonance Imaging (MRI) . . . . . . . . . . . . . 20
3.1.2 Computed Tomography (CT) . . . . . . . . . . . . . . . . . 24
3.2 Medical Image Watermarking . . . . . . . . . . . . . . . . . . . . . 27
4 Proposed System for Medical Image Watermarking preserving
ROI 30
4.1 Separating ROI from medical image . . . . . . . . . . . . . . . . . . 31
4.2 Medical Image Watermarking System . . . . . . . . . . . . . . . . . 33
4.2.1 Integer to Integer transform . . . . . . . . . . . . . . . . . . 34
4.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3.1 Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5 Experiments and Results 41
5.1 The experiments and results of the system without attacks . . . . . 41
5.1.1 CT Scan Images . . . . . . . . . . . . . . . . . . . . . . . . 41
5.1.2 MRI Images . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.1.3 X-Ray Images . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.4 Ultrasound Images . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 Embedded and extracted watermark with attacks . . . . . . . . . . 46
6 Conclusion and Future Work 53
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Bibliography 55
vi
Page 8
List of Figures
1.1 A Typical Watermarking System . . . . . . . . . . . . . . . . . . . 3
1.2 Types of Watermarking Techniques . . . . . . . . . . . . . . . . . . 9
1.3 The filter bank structure used in wavelet decomposition of an image 11
2.1 A typical e-diagnosis Model . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Block diagram of Medical Image watermarking . . . . . . . . . . . . 16
3.1 Medical image indicating ROI . . . . . . . . . . . . . . . . . . . . . 20
4.1 Medical Image Watermarking Approach Preserving ROI . . . . . . 31
4.2 Interface for GUI based approach . . . . . . . . . . . . . . . . . . . 32
4.3 Sub-band structure a of 4-level wavelet transform . . . . . . . . . . 36
4.4 Ultrasound Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5 Quantization Procedure . . . . . . . . . . . . . . . . . . . . . . . . 38
5.1 Segmenatted ROI of host image . . . . . . . . . . . . . . . . . . . . 42
5.2 (a) The original host CT scan image, (b)Roi removed image,(c)Emebedded
image without ROI, (d)Final embedded image with ROI . . . . . . 42
5.3 Recovered original image . . . . . . . . . . . . . . . . . . . . . . . . 42
5.4 Embedded and extracted watermark values without any attacks . . 43
5.5 (a) The original host MRI image, (b)ROI image of host image,(c)Roi
removed image (c)Emebedded image without ROI, (d)Final embed-
ded image with ROI . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.6 Embedded and extracted watermark values without any attacks for
MRI image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.7 (a) The original host X-Ray image, (b)ROI image of host image,(c)Roi
removed image (c)Emebedded image without ROI, (d)Final embed-
ded image with ROI . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.8 Embedded and extracted watermark values without any attacks for
X-Ray image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Page 9
5.9 (a) The original host Ultrasound image, (b)ROI image of host
image,(c)Roi removed image (c)Emebedded image without ROI,
(d)Final embedded image with ROI . . . . . . . . . . . . . . . . . . 46
5.10 Embedded and extracted watermark values without any attacks for
Ultrasound image . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.11 (a) The original watermarked CT scan image, (b)The image after
sharpning attack with 0.02 factor . . . . . . . . . . . . . . . . . . . 48
5.12 Embedded and extracted watermark values with sharpning attck
(0.02 factor) image . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.13 (a) The original watermarked MRI image, (b)The image after His-
togram attack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.14 Embedded and extracted watermark values after histogram attack
on image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.15 (a) The original watermarked X-Ray image, (b)The image after
10% JEPG compression attack . . . . . . . . . . . . . . . . . . . . . 50
5.16 Embedded and extracted watermark values after JEPG Compres-
sion attack on X-Ray image . . . . . . . . . . . . . . . . . . . . . . 50
5.17 (a) The original watermarked Ultrasound image, (b)The image after
up and down sampling attack . . . . . . . . . . . . . . . . . . . . . 51
5.18 Embedded and extracted watermark values after down and up sam-
pling attack on Ultrasound image . . . . . . . . . . . . . . . . . . . 52
viii
Page 10
Chapter 1
Introduction: Digital
Watermarking
1.1 Digital watermarking
In recent year all the business applications are moving towards the digital era,
because of great development in latest technologies such as in the area of com-
munication, networked multimedia system, digital data storage etc. Also from
the last two decades use of internet is rapidly increased in business environment
towards achievement of effectiveness, convenience and Security by introducing the
digitization in their work.
It was estimated that in 1993 the Internet will carry only 1% of the informa-
tion however by 2000 this figure had grown to 51%, and by 2007 more than 97 %
information was carried away across the globe. A study conducted by JupiterRe-
search says that 1.1 billion people have regular Web access and use application like
electronic mail, instant messaging, social networking, online messaging etc. which,
helps in growth & knowledge sharing in different domains such as education, re-
search, development, Medical, and many business etc. In business applications to
speed up the business process communication use of digital media has been drasti-
cally increased. This digital data includes text, images, audio, video and software
which are transferred over open public network, hence there is need to protect this
data. There are many techniques that are available for protection of this digital
data, such as encryption (cryptography), authentication and time stamping. Also
there is another method that improved the protection of digital data by merging
a low level signal directly into the digital data. This low level signal is known as
watermark, that uniquely identifies the ownership and provide the security to the
1
Page 11
1.2 Principle of Digital Watermarking
digital data and can be easily extracted.
The process of embedding the watermark into a digital data is known as Dig-
ital Watermarking. It is a process of embedding unremarkable logos or labels or
information data or pattern into the digital data [1]. The concept of digital wa-
termarking is associated with the stegnography. It is defined as covered writing,
which hides the important message in a covered media while, digital watermark-
ing is a way of hiding a secret or personal message to provide copyrights and the
data integrity. Digital image watermarking is a new approach, which is suitable
for medical, military, and archival based applications. The embedded watermarks
are difficult to remove and typically imperceptible, could be in the form of text,
image, audio, or video.
The embedding of secret watermark in digital data, no matter how much invis-
ible it may be. However it leads to some degradation in the resultant embedded
digital data. To overcome this and to retrieve the original data, reversible wa-
termarking has been implemented, which considered as a best approach over the
cryptography. In cryptography after encryption the resultant data may not be
visible or understandable also at the time of retrieval this may lead to loss of se-
mantic information of host data, which is not in case of watermarking. In digital
data several watermarks can be embedded at the same time and this is known as
multiple watermarking technique. A digital watermark also considered as digital
signature which provides the authenticity. A given watermark may be unique to
each copy (e.g. to identify the intended recipient), or be common to multiple
copies (e.g. to identify the document source).
1.2 Principle of Digital Watermarking
Basically, digital watermarking is consisted of two main processes, namely embed-
ding process and extracting process. During the embedding process, watermark
is embedded into the multimedia data (digital data). The original digital data
(multimedia content) will slightly modified after embedding the watermark, this
modified data is called as watermarked data. While in extraction process this
embedded watermark is extracted from the watermarked data and recovers the
original multimedia data. The extracted watermark is then compared with the
original watermark; if the watermark is same it results in authenticated data.
During the transmission of the watermarked data over the public network at-
2
Page 12
1.2 Principle of Digital Watermarking
tacker may tamper the data, and if any modification in the data can be detected
by comparing the extracted watermark with the original watermark.
Figure 1.1: A Typical Watermarking System
A typical watermarking system is shown in Figure 1.1 which includes water-
mark embedder and watermark extractor. The inputs to the embedder are multi-
media data and watermark, which is to be embedded into the original multimedia
data. The output of watermark embedder is watermarked data (watermarked
content). The inputs to the watermark extractor depending on the method are
original multimedia or original watermark. The watermark extraction process in-
volves two steps [2]. In the first step one or more pre-process is applied on the
watermarked data to extract a vector called extracted watermark. Then the sec-
ond step is to determine whether the extracted watermark is same as original
watermark by comparing the extracted watermark with the original watermark
called reference watermark. The result of second step is to measure the confidence
by indicating how likely the original watermark is present in the digital data [3].
The multimedia data in Figure 1.1 includes text [4], image [5], audio file [6], video
[7], 3D data [8, 9], and object [10].
Suppose that X is the original multimedia data and W is the watermark to
be embed. In digital watermarking system a embedding function E(.) takes X
and W as a input values and gives X ′ i.e. watermarked data as a output. X ′ is
3
Page 13
1.3 Types of Watermarking System
obtained as:
X ′ = p (X,W ) (1.1)
The embedding algorithm is considered as robust if watermark is embedded in a
way such that it can survive even if the watermarked data X’ goes through several
attacks. During extraction process the extraction function D(.) is defined as:
W ′ = D (X ′, [X], [W ]) (1.2)
Where W’ is retrieved watermark, X and W enclosed in braces [ ]can be optional
inputs for extraction function, which depends on the application. For example
[X] is used when the watermarking system is non-blind, this system is suitable for
the application where to extract the watermark original image is needed. If the
watermarking system is blind the input to the extraction function is [W] only.
A typical watermarking should satisfy the following requirements.
• The watermark W should be extracted from X’ with or without X
• X’ should be as close to X as possible
• If X’ is not manipulated/modified, the extracted watermark should be same
as W
• For robust watermarking, if X’ is modified, W’ should still match W to give
clear judgment of the existence of watermark
• For fragile watermarking, after even the slight manipulation to X’ extracted
W’ should be totally different from W. In such system W indicates the
tampering to the X’
1.3 Types of Watermarking System
Depending on the application, watermarking system can be of different types.
1.3.1 Visible watermarking system
In visible watermarking system watermark (text or image) is semi-transparently
embedded into original data. Visible watermarking is more robust against image
4
Page 14
1.3 Types of Watermarking System
transformation attacks, which provides copyrights protection of intellectual prop-
erty thats in digital format. In visible watermarking watermarked data is view as
digitally stamped document.
1.3.2 Invisible watermarking system
In invisible watermarking system watermark is embedded into the original data
in such a way that the embedded watermark should not be visible by naked eyes.
Only electronic devices (or specialized software) can extract the embedded in-
formation to prove the authenticity. Such type of system is used to identifying
the source, author, creator, owner, and distributor or authorized consumer of a
multimedia data.
1.3.3 Blind watermarking system
A watermarking technique is said to be blind, if to extract the watermark from
watermarked data it does not need original image. The blind watermarking system
is also known as oblivious. Blind watermarking system is more popular because
it decreases the overhead of cost and memory for storing original data.
1.3.4 Non-blind watermarking system
The watermarking techniques in which to extract the watermark, it requires the
original data is known as non-blind watermarking system. It is more robust than
blind watermarking system.
1.3.5 Robust watermarking system
A watermarking system is said to be robust, if any modification on the water-
marked data results in no change into watermark value. That is extracted water-
mark information from the tampered watermarked data would be same as original
watermark information. A robust watermarking system resist against wide range
of intentional and unintentional attacks such as, image enhancement, filtering,
noise addition, JPEG compression and geometrical transformations, collusion and
forgery attacks.
Robust watermarking systems have been proposed to be implemented in num-
ber of application. Such as copyright protection, finger printing and access control.
5
Page 15
1.4 Properties of Digital Watermarking
Copyright protection is one of the main applications of robust watermarking sys-
tem. In copyright protection application the idea is to embed information about
the copyright owner into the multimedia data to prevent parties from claiming to
be the rightful owners of the data. The robust watermark embedded into the con-
tent is detectable despite common image processing manipulations. finger printing
is used to trace authorized users who violate the license agreement and distribute
the copyrighted material illegally. Thus, the information embedded in the content
is usually about the customer such as customer’s identification number.
1.3.6 Fragile watermarking system
In fragile watermarking system embedded watermark in host data can be eas-
ily destroyed. This property is useful to identify whether a multimedia data is
modified/ manipulated or not? By embedding then fragile watermark into mul-
timedia data, the authenticity of multimedia data can be achieved. Any small
manipulation on the watermarked data will lead to distortion in corresponding
embedded fragile watermark. At the end side by comparing the extracted water-
mark with original watermark, it can be easily identified whether the multimedia
data is manipulated or not. The different applications where fragile watermark-
ing can be used are document authentication, evidence authentication, complete
authentication etc.
1.4 Properties of Digital Watermarking
An effective digital watermarking algorithm must have number of properties. This
section describes the number properties of digital watermarking algorithm.
1.4.1 Imperceptibility
The basic requirement of digital watermarking is to have the watermarked im-
age should look alike as the original image. This confirms there is not much
degradation on the original image. This property is known as imperceptibility or
transparency of the watermarking system [11]. The embedded watermark should
not be visible to human eye. To calculate the imperceptibility, generally Peak
Signal to Noise Ratio (PSNR) is used [11].
6
Page 16
1.4 Properties of Digital Watermarking
1.4.2 Robustness
The capability of survival of watermark against both legitimate and illegitimate
attacks is referred as robustness. All watermarking system needs to resists against
any legitimate and illegitimate attacks, except fragile watermarking system. For
manipulation recognition in original data the watermark has to be fragile to detect
altered media. Robustness depends on watermarks information capacity, visibility
and strength. Generally a good watermarking algorithm should be robust against
filter processing, noise addition, geometrical transformations such as rotation, scal-
ing, translation and lossy compression such as JPEG compression [12].
1.4.3 Security
The watermarking system should be secured i.e. hacker should not be in position
to extract the watermark without having the knowledge of embedding algorithm.
Watermarking system must be capable of stand firm against different attacks
[2]. Attacks try to remove, modify or embed (unwanted information) into the
watermark. Attacks are mainly classified in two different types i.e. passive attack
and active attack. Passive attack only detects the watermark information, while
active attack tries to modify the watermark information.
1.4.4 Complexity
The time and effort needed to embed and retrieve the watermark information
is known as complexity of the watermarking system. The complex algorithm in
watermarking system requires more software and hardware resources to implement
it, which results in increasing the computation cost. To reduce the computational
cost of watermarking system, it should be less complex. Such as in telemedicine
domain, to cut the cost of bandwidth consumption during the transmission of
medical data less complex watermarking algorithms are implemented.
1.4.5 Capacity
Capacity of the watermarking system describes embedding of maximum amount
of watermark information i.e. embedding the multiple watermarks in single data.
The higher capacity of embedding information in a data can be obtained by com-
promising either imperceptibility or robustness of algorithm [13].
7
Page 17
1.5 Watermarking Techniques
1.4.6 Invertibility
This property of digital watermarking system describes the possibility of generat-
ing original data during the extraction process of watermark.
1.4.7 Verification
This property defines the procedure of verification i.e. private key verification and
public key verification, depending on its respective algorithm.
1.5 Watermarking Techniques
There are different kinds of watermarking techniques are in place, which are differ-
entiated on the basis of types of document, types of domain, etc [14]. The various
types of watermarking according to different categories are shown in Figure 1.2.
Watermarking techniques are broadly divided into four types:
1. According to working domain
2. According to types of document
3. According to human perception
4. According to application
These four categories are further classified as below
1. According to types of document
• Text watermarking
• Image watermarking
• Audio watermarking
• Video watermarking
2. According to human perception
• Visible watermarking
• Invisible watermarking
8
Page 18
1.5 Watermarking Techniques
Figure 1.2: Types of Watermarking Techniques
3. According to Application
Source based watermarking: This approach is used for the ownership au-
thentication where unique watermark is embedded into all copies of data.
Destination based watermarking: This approach is used in the application
where the tracing of buyer is done for the purpose of illegal reselling. Here
for each distributed copy a unique watermark is used.
4. According to working domain
• Spatial domain
• Frequency domain
Watermark can be applied in spatial domain or it can be applied in frequency
domain.
1.5.1 Spatial domain watermarking
Spatial domain watermarking method hides the watermark directly within the host
data [15, 16]. This approach is easy and simple to implement. The advantage of
this approach is the spatial localization of the embedded data can be achieved
automatically even after the watermarked content goes under some attacks. An-
other advantage of spatial domain watermarking is that, it allows the control on
maximum difference between the original image and watermarked image due to
which design of near-lossless system can be possible [13]. Spatial domain water-
marking is applied in number application. There are various ways of applying
9
Page 19
1.5 Watermarking Techniques
spatial domain watermarking.
(a) Additive Watermarking
Additive watermarking is most straightforward method for embedding the water-
mark in spatial domain. It adds pseudo random noise pattern to the pixel of host
data. To ensure that embedded watermark should be detected, the noise to add
in host data is generated by a key. The same key is used at the time of extraction
process.
(b) Least Significant Bit Modification
This method is very common for embedding the watermark in the host data. It
relies on the way of manipulating the LSBs of host data, in a manner which is not
detectable by human eye. The basic idea for this method is to replace LSBs of
host data by same size of binary watermark.
1.5.2 Frequency domain watermarking
However, the spatial-domain watermark insertion is simple and easy to implement,
but it is fragile versus various attacks and noise. To get the better robustness as
well as imperceptibility, watermarking is done in frequency domain. Frequency do-
main is also known as multiplicative watermarking. There are several watermark-
ing techniques in different frequency domain such as Discrete Fourier Transform
(DFT), Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT),
Discrete Curvelet Transform, and Discrete Counterlet Transformation [17]. This
section covers the details of DWT domain.
Discrete Wavelet Transform (DWT)
All transform domain watermarking algorithms generally follows three steps i.e.
(i) Data transform (ii) watermark embedding and (iii) Watermark recovery. Trans-
formation of host data can be applied either on whole data [18], or in block by
block manner [19]. Wavelets are mathematical function that cuts the data into
different frequency components, and according to the resolution matched to its
scale wavelet function study each component. The advantage of wavelet trans-
form over traditional Fourier methods is that it analyses the signal which contains
discontinuities and sharp spikes. Other advantage of wavelet transform is, it cap-
tures both frequency and location information (location in time). The basic idea
for 1D DWT is it decomposes the signal (host data) into high frequency part and
low frequency part. The edge components of the signal are largely contained to
10
Page 20
1.5 Watermarking Techniques
the high frequency part. The low frequency part is again split into two parts low
and high frequency part. This process can be continued till an arbitrary number,
which is usually determined by the application at hand. Furthermore, the original
signal can be reconstructed by inverse DWT (IDWT) process. In case of 2D-DWT
we get four subbands from one level, that are Low-Low level (LL), High-High Level
(HH, Low-High level (LH) and High-Low level (HL). The LL subband contains the
low level details of the image. In the next level, the 2D-DWT of the LL subband
is obtained and this is repeated in each succeeding level.The filter bank structure
used in wavelet decomposition of an image is shown in Figure 1.3. Where h[n] is
high pass filter, g[n] is low pass filter and W is wavelet function.
Figure 1.3: The filter bank structure used in wavelet decomposition of an image
H[n] and G[n]are defined as below:
H[n] =∑k
hke−jkn (1.3)
11
Page 21
1.6 Application of Watermarking
G[n] =∑k
gke−jkn (1.4)
There are number of basic function that can be used to perform wavelet transform
on given signal. Such as, lazy, haar, daubechies wavelets (db1, db2, db3, etc.),
meyer, etc.
1.6 Application of Watermarking
Increasing research on watermarking from the past decades has been largely mo-
tivated by its applications in copyright management and protection. The digital
watermarking technique is highly suitable for medical, military, and archival based
applications.
• Broadcast monitoring is the well known application of watermarking, which
helps advertising agencies to track the specific video broadcast by a TV
Channel or station. Embedding the watermarked video to the host video
will provide you easier way to track and monitor the broadcast.
• Owner Identification is also a well known application of watermarking, which
helps in identifying the owner of video or image. Such as copyright authori-
ties, where instead of providing copyright notice with every image or video
the watermark could be directly embedded in to the image or video itself.
• Another well know application of watermarking is copy control which helps
preventing the illegal copy of songs or images of movies etc. Where by
embedding watermark in songs or images of movie would instruct a water-
marking compatible DVD or CD writer to not write the song or movie as it
is an illegal copy.
• With the help of watermarking Transaction Tracking can be achieved by
recording the transaction details in the history of a copy in digital work.
For example issuing each recipient a legal copy of movie by embedding the
watermark (different watermark for different recipient) will help in tracking
the source of leak in case of movie leaked to the internet.
• Medical image watermarking is one of the important applications of wa-
termarking. Medical image authentication systems which can not only au-
thenticate medical images but would also be able to secretly communicate
auxiliary information can be achieved by watermarking technique. Only the
12
Page 22
1.7 Thesis Outline
authorized people of the hospital would thus be able to modify the content
of medical image.
1.7 Thesis Outline
The outline of report is described below:
Chapter 2 provides brief introduction to the medical image watermarking, where
it explains the requirements of medical image watermarking. The study of differ-
ent segmentation algorithm in place, to separate the ROI from medical image and
the available algorithm for medical image watermarking are discussed in chapter 3.
Chapter 4 discuss about the proposed system for medical image watermarking
preserving ROI. The proposed system has been applied successfully against all
existing medical imaging. Chapter 5 shows the experimental results achieved by
using the proposed system. Chapter 6 provides insight on conclusion and the
future work.
13
Page 23
Chapter 2
Medical Image Watermarking
Introduction
Speedy development of internet in every field leads to availability of digital data
to the public. Internet has been spread in many applications like telemedicine,
online-banking, teleshopping etc. One of this application telemedicine is crucial
one, where Internet is used to transfer or receive medical data by healthcare pro-
fessional. Due to advancement in information and communication technologies, a
new context of easier access, manipulation, and distribution of this digital data
have been established [20]. The medical images can be readily shared via com-
puter networks and easily used, processed, and transmitted by using great spread
network [21, 22].
In the last decades, uses of advanced electronic and digital equipments in
health care services are increased, where traditional diagnosis system has been re-
placed by e-diagnosis system. In fact, in most of the hospitals physicians diagnose
their patients by relying on the provided electronic and digital data (such as Ul-
trasonic, Computed Tomography (CT), Magnetic Resonance Imaging (MRI) and
X-ray images). This results in generation of large number of electro digital data
(i.e. medical images) continuously at various health care centers and hospitals
around the world. The typical e-diagnosis model is shown in Figure 2.1, where
medical image can be sent by patient through the internet to physician. One
physician can transfer the medical image to anther physician for second opinion.
The medical images are stored in patient historical database for future diagnosis.
In number of medical applications, special safety and confidentiality is required
for medical images, because critical judgment is done on medical images, which
14
Page 24
leads to the proper treatment. Therefore, it must not be changed in an illegitimate
way; otherwise, an undesirable outcome may results due to loss of decisive infor-
mation. Therefore, there is a need to provide a strict security in medical images to
ensure only occurrence of legitimate changes. Now-a-days exchange of medical im-
ages between hospitals located in different geographical location is very common.
Moreover, as this exchange of medical reference data done via unsecured open
networks leads to the condition of changes to occur in medical images and creates
a threat which results in undesirable outcome. Considering this fact, demand of
security is getting higher due to easy reproduction of digitally created medical
images. For copyright protection and authentication of these medical images, dig-
Figure 2.1: A typical e-diagnosis Model
ital watermarking is an emerging technique, which includes the embedding and
extraction process. Embedding process hides some secrete information in to med-
ical images. This secret information is extracted during the extraction process. If
failure occurs in extraction process the physician would come to know that there
has been some kind of tampering with that image, and he would take precaution
of not making diagnosis based on that image. However, if the extraction process
extracts the correct watermark, which generally consumes a few seconds, physician
can continue with diagnosis.
15
Page 25
2.1 Principle of Medical Image Watermarking
2.1 Principle of Medical Image Watermarking
The typical block diagram for medical image watermarking is given in Figure 2.2.
Encoder E embeds the watermark W in medical image to provide security and
authentication. Decoder D extracts the watermark from watermarked image. By
comparing the extracted watermark with original watermark, one can affirm the
tampering of medical image.
Figure 2.2: Block diagram of Medical Image watermarking
To ensure the reliability and quality of the watermarked image, the performance
of watermarking is calculated, which measured in terms of perceptibility. There
are two method of calculating the performance measure.
• Mean Square Error (MSE):
It is simplest function to measure the perceptual distance between water-
marked and original image. MSE can be defined as:
MSE =1
n
n∑i
(I ′ − I)2
(2.1)
Where, I is original image and I is watermarked image.
• Peak Signal to Noise Ratio (PSNR):
It is used to measure the similarity between images before and after water-
marking.
PSNR = 10 log10
maxI2
MSE(2.2)
Where, max I is the peak value of original image.
16
Page 26
2.2 Requirements of Medical Image Watermarking
2.2 Requirements of Medical Image Watermark-
ing
2.2.1 Imperceptibility
Imperceptibility is one of the strict requirements of the medical image watermark-
ing. Imperceptibility means the embedded watermark should not be visible by
human eye. It is often not allowed to alter the medical image contents even after
embedding the watermark in some application [23, 20]. The imperceptibility in
medical image watermarking can be achieved by two methods. In first method
imperceptibility is fulfilled by selecting the Region of Non Interest watermarking
[24], in which the watermark is embedded in RONI area of medical image. In
this method the Region of Interest (ROI) area of medical image will be distortion
free. Imperceptibility can be achieved by reversible watermarking method which
recovers the original medical image by undoing the watermark embedding process
at the receiving side [25].
2.2.2 Capacity
In medical image watermarking, all the information that are required by the physi-
cian such as identification of patient, doctor identification, treatment, etc are em-
bedded in medical image. Therefore, capacity for embedding the payload must be
high.
2.2.3 Authenticity
The entitled users (patient, doctor) should have the access to the medical data.
This can be achieved by embedding the doctors identification and patient identi-
fication in medical images.
2.2.4 Reversibility
At the receiving side the reverse of embedding process should be possible to get
the original medical image and embedded watermark. This property is known as
reversibility of medical image watermarking.
17
Page 27
2.2 Requirements of Medical Image Watermarking
2.2.5 Complexity
According to the e-diagnosis the medical images are transferred from some remote
location to the other side through internet. In such cases speed becomes an im-
portant matter, thus the algorithm should be less complex to reduce the execution
time.
2.2.6 Intactness of ROI
Medical images hold decisive property and are very crucial and important part
of medical information. Such part of the medical image is called as Region of
Interest (ROI). The ROI is helpful in providing further diagnosis by the physician.
A small bit of distortion in ROI may lead to undesirable treatment for patient. For
securing medical images through watermarking ROI should be preserved and the
watermarks can be applied on the remaining part of the image called as Region of
Non Interest (RONI). Therefore, application of watermarking in medical images
can be considered as two-step process which includes:
1. Extracting ROI form the medical images
2. Applying watermarking on RONI
18
Page 28
Chapter 3
Literature Survey
Different algorithms are available for segmentation of ROI on the different types of
medical images. Additionally, there are different algorithms available for applying
watermarking.
3.1 Region of Interest (ROI) Segmentation
Segmentation plays an important role in medical image processing [26, 27]. In
medical image analysis segmentation is the first step to be followed, to avoid dis-
tortion of ROI [26, 28]. Image segmentation deals with the process of partitioning
an image into different regions by grouping together neighborhood pixels based on
some predefined similarity criterion [29]. This similarity criterion can be defined
by specific properties of pixels in the image. Segmentation in medical imaging
is used for the extracting the features, image display and for the measurement of
image. The goal of segmentation is to divide entire medical image in to sub regions
i.e. (white and gray matter). In addition, this helps in classifying image pixels
in to anatomical regions (such as bones, muscles and blood vessels). Defining
the borders of ROI in medical image can simplify the procedure of segmentation.
In addition, the step of defining borders of ROI is a crucial one, which helps
in determining the result of the application as entire analysis fully relies on the
output from segmentation step. There are different approaches (for segmenting
the image) defined for the different imaging technologies such as CT, MRI, US,
colonoscopy etc. Segmentation is semi-automatic procedure and we need to define
a seed point in an image. Therefore, the algorithm, which gives perfect result for
one application, might not even work for another. Figure 3.1 shows the ROI part
of medical image, where physician performs the diagnosis.
We have various existing medical imaging like Computed Tomography (CT),
19
Page 29
3.1 Region of Interest (ROI) Segmentation
Figure 3.1: Medical image indicating ROI
Magnetic Resonance Imaging (MRI), Ultrasound (US), and Positron Emission
Tomography (PET) etc. Here, two most common imaging i.e. MRI and CT scan
are discussed in detail with their proposed algorithms.
3.1.1 Magnetic Resonance Imaging (MRI)
Magnetic Resonance Imaging (MRI) provides a wealth of information, which is
useful for medical examination. In many applications where MRI is used, segmen-
tation of image into different intensity classes are needed, which is regarded as the
best available representation for biological tissues [30, 31]. Segmentation plays
very important role in MRI process for deciding the spatial location, selecting the
operation path, shape, and size of the focus etc. In segmenting MRI images, main
requirement is to care about three problems: noise, partial volume effects (where
more than one tissue is inside a pixel volume), and intensity in-homogeneity [32].
Due to irregularities of the scanner magnetic fields-static (BO), radio frequency
(B1), and gradient fields, intensity in-homogeneities are caused. These irregu-
larities results in producing spatial changes in static tissues of MRI data. When
multiple tissues contribute to a single voxel, by making the distinction between tis-
sues along boundaries more difficult leads to the problem of partial volume effects.
Adding noise in MRI images can encourage disconnection between segmentation
regions. Therefore, for doing segmentation of MRI data on these three difficulties
need to focus. There are four different approaches for doing image segmentation:
20
Page 30
3.1 Region of Interest (ROI) Segmentation
thresholding, clustering, edge detection, and region extraction. This section covers
various available MRI segmentation algorithms based on following approaches:
• Thresholding approach
• Clustering approach
• Edge detection approach
1. Thresholding: Thresholding is one of the easiest and most frequently used
techniques to segment MRI data by separating the foreground from back-
ground of image [33, 34]. Thresholding approach can further be classified as
global thresholding and local (adaptive) thresholding. In global thresholding
method, image segmentation is done by providing single threshold value in
the whole image whereas, in local thresholding, threshold value is assigned to
each pixel of image by using local information around the pixel, and then to
determine whether the particular pixel belongs to foreground or background
these threshold values are used. Due to simplicity and easy implementation
of global thresholding, this method is more popular. P-tile method is one of
the earliest thresholding methods based on the gray level histogram [33, 35].
Here P refers to the word percentile. This algorithm stands on the statement
Objects in the image are brighter than the background, which occupy a fixed
percentage of the picture area. In this algorithm, threshold is defined as the
gray level that mostly corresponds to mapping at least P% of the gray level
into the object. The experimental results of this method specify that it is
suitable for all size of objects, and it provides good anti-noise capabilities.
However, this method is not applicable in application where object area ratio
is unknown or varies from image to image.
2. Clustering: The goal of clustering approach is to group similar objects and
separates the dissimilar objects. That is depending on some perceived simi-
larities this grouping of pixels is done. These clusters then lead in providing
natural partitions of pixels that corresponds to the different regions in an
image. Conventional clustering algorithms require a prior knowledge regard-
ing the number of clusters, clustering criteria, and nature of data, etc. There
are many algorithms defined for the clustering, such as K-means clustering,
Fuzzy c-means (FCM) clustering, possibilistic c-means, possibilistic-fuzzy
clustering, intuitive fuzzy c-means (IFCM), and so on. The objective of clus-
tering, for a given set of unlabeled N samples or data i.e. X= x1, x2, ...., xn
21
Page 31
3.1 Region of Interest (ROI) Segmentation
is to assign a class label among C labels to each of N samples. This num-
ber of labels C, considered as the number of regions or number of groups.
One of the most widely used clustering algorithms is Fuzzy c- means (FCM)
[36, 37]. In the FCM algorithm, it assigns labels to the data, which are in-
versely related to relative distance of to the point prototypes that are cluster
centers in FCM model. In FCM, proximity of each data or samples, xk, to
the center of cluster, vi, is defined as membership or label (uki) of data xk
to the ith cluster of X with following conditions:
0 ≤ uki ≤ 1andc∑
i=1
uki = 1,∀k (3.1)
U = [uki]NxC andV = v1, v2, ....vi (3.2)
FCM algorithm is acknowledged as one of the best clustering algorithm as it
resolves various problems, [36, 38], although it still suffers from undesirable
solutions with outliers data [36, 39]. As in FCM algorithm, it requires to
provide the exact number of clusters in advance as a prior knowledge. The
exact estimation of number of clusters in MRI images, (used in particular
for diseased cases), may not be possible to have in advance. To overcome
this problem, Krishnapuram and Keller proposed a clustering model named
possibilistic c-means (PCM) [36, 38]. In the PCM model the condition of
FCM model,
C∑i=1
uki = 1 ∀k, is relaxed by introducing the new condition as (3.3)
C∑i=1
uki = C ∀k (3.4)
By providing new condition, PCM model improves its performance over the
FCM algorithm, as PCM overcomes the drawback of FCM. PCM is ISO-
DATA based algorithm, which makes use of user defined criteria for merging
and splitting clusters to discover the number of natural clusters in the data.
However, it is very difficult to define these splitting and merging criteria that
can be applied on various MR data based on prior assumptions of intensity
distribution. Hence, PCM model is very sensitive to initialization and need
22
Page 32
3.1 Region of Interest (ROI) Segmentation
of additional parameters [36, 40]. Afterward a possibilistic-fuzzy clustering
(PFCM) model is proposed. In PFCM algorithm both FCM and PCM model
combined by introducing two new parameters a and b, where a and b are
the weighting factors of FCM and PCM, respectively, to resolve the outlier
problems of FCM and sensitivity problem. Again, in PFCM algorithm as it
requires selecting additional parameters and extra computation complexity,
a new model intuitive fuzzy c-means (IFCM) model is proposed by Dong-
Chul Park for MRI image segmentation. The basic operation of IFCM model
is same as in FCM except the membership assignment condition. To deal
with the problems of membership assignment to noise data IFCM algorithm
has been developed. In IFCM, a new measurement called intuition level is
introduced by using membership values of FCM and PFCM, uki so that the
intuition levels may alleviate the effect of noise data.
3. Edge Detection: For doing the segmentation by using edge detection ap-
proach, first step is to extracts the features by obtaining the information
from images. Edge detection is a fundamental tool used in most image
processing application. It is the process of detecting boundaries between
objects and the background in the image, at which the image brightness
changes sharply. There are many algorithms to perform edge detection, and
all of them classified into two categories Gradient and Laplacian.
Edge detection based on Gradient method initially calculates first derivative
of image, and then find its corresponding local maxima and minima values
to detect the edges. While, in the Laplacian method after obtaining the
second derivative of the image it looks for zero crossing. There are various
operators defined i.e. Roberts [41], Prewitt [42], Sobel [43], Canny [44] edge
operators to perform Gradient method. These operators include a small ker-
nel rolled up together with the image, which helps in estimating first order
directional derivative of the image brightness distribution. This kernel finds
edge strength in the direction, which are orthogonal to each other, usually
vertically and horizontally. The total value of the edge strength is then
obtained by the combination of both the components. Here by creating a
matrix centered on each pixel it calculates the edge value. Moreover, if the
calculated value is larger than provided threshold, then that pixel is classi-
fied as an edge.
23
Page 33
3.1 Region of Interest (ROI) Segmentation
With reference to the earlier work of the Marr and Hilderth [42], John F.
Canny [44] has developed an edge detection operator named as Canny edge
detection operator in 1986. This operator helps in detecting wide range
of edged in images by using a multi-stage algorithm. He provided gradient-
based-finding algorithm called as optimal edge detector, which becomes most
popular and commonly used edge detectors to have the segmented image.
Canny edge detector used a method called Hysteresis, which is proposed to
tracing the unsuppressed pixels. Here it uses two threshold values i.e. high
and low. After finding the gradient values, algorithm compares these values
with the provided two threshold values. The pixel is set to zero if gradient
value is below the low threshold value and if it is above the high threshold
value then pixel is set as an edge. In case if, the gradient value is in between
the two threshold values by default that pixel is set to zero (regarded as non-
edge) although there is a path from that pixel to the pixel having gradient
value above the high threshold value.
3.1.2 Computed Tomography (CT)
Computed Tomography (CT) scanning sometimes called Computed Axial Tomog-
raphy (CAT) scanning [29], is a noninvasive medical test that helps physicians
diagnose and treat medical conditions. CT scanning combines special x-ray equip-
ment with sophisticated computers to produce multiple images or pictures of the
inside of the body. These cross-sectional images of the area being studied can then
be examined on a computer monitor, printed or transferred to a CD. CT scans
of internal organs, bones, soft tissue and blood vessels provide greater clarity and
reveal more details than regular x-ray exams. Using specialized equipment and
expertise to create and interpret CT scans of the body, radiologists can more eas-
ily diagnose problems such as cancers, cardiovascular disease, infectious disease,
appendicitis, trauma, and musculoskeletal disorders. Hence, the CT scan appli-
cation is been widely used in medical domain. There are different segmentation
methods proposed considering CT scan of different body organs (such as lung,
liver, kidney, etc.) This section covers the different segmentation algorithm for
CT scan images for protecting the distortion of diagnosis value.
The 2-D and 3-D segmentation of organs in medical application of image pro-
cessing are classified into model based and nonmodel based approaches. Nonmodel
based approaches depends on local information such as, texture, intensity, spatial
24
Page 34
3.1 Region of Interest (ROI) Segmentation
correlation of 2-D organ image in consecutive slices, and the location of the organ
in the abdominal area with respect to neighboring structures, e.g., spine and ribs
[45]. Various segmentation algorithms are developed using nonmodel-based ap-
proach. This section first covers the different segmentation algorithm, which uses
nonmodel-based approach. Susomboon et al. [46] presented texture features to
perform region classication for extracting livers soft tissue. Seo et al. [47] employed
a multimodal threshold method based on piecewise linear interpolation that used
spine location as a reference point. Forouzan et al. [48] introduced a multilayer
threshold technique, in which by statistical analysis of the liver intensity it cal-
culates the threshold value. Both these methods use the local information of the
livers relative position to the spine and ribs. Nonmodel-based methods for organ
segmentation leads to inaccuracies due to variation in imaging condition, because
of occurrence of tumor inside the organ and noise. Dependencies on prior infor-
mation such as texture and image values could cause inaccuracies in segmentation
process as feature could change from one patient to another. Moreover, most of
these methods are parameter dependent and hence for the best performance it
often needs to adjust the parameters from one CT volume to other. In recent
years, model-based image segmentation algorithms developed for various medical
applications. These methods aim to recover an organ based on statistical informa-
tion. State-of-the-art algorithms on model-based segmentation are based on active
shape and appearance models [45]. Model-based techniques provide more accu-
rate and robust algorithm for segmenting the CT scan image. These techniques
also deal with the missing image features via interpolation. The performance of
these methods depends on the number and type of training data. Moreover, if
the shape to be segmented lies too far from the model space, that might not be
detected by many those better methods which does not implemented by statistical
model-based approach.
Pan and Dawant [49] reported a geometrical-level set method for automatic
segmentation of the liver in abdominal CT scans without relying on the prior
knowledge of shape and size. Even if this method depends on a model-based
technique, that outperforms threshold-based techniques, but it did not use prior
knowledge of the liver shape. Lin et al. [51] presented the algorithm to perform
segmentation of kidney, based on an adaptive region growing and an elliptical
kidney region positioning that used spines as landmark. H. Badakhshannoory
and P. Saeedi [51] incorporated a method for liver segmentation. Based on liver
boundary edges to identify liver regions, nonrigid registration and a multilayer
25
Page 35
3.1 Region of Interest (ROI) Segmentation
segmentation technique are combined in this approach. This method is does not
affected by the diversity of existing liver shapes, as it does not rely on any shape
model. Samuel et al. [52] has proposed the use of Ball-Algorithm for the seg-
mentation of lungs. In this algorithm at the first stage, it applies the grey level
thresholding to the CT images to segment the thorax from background and then
the lungs from the thorax. Then in the next step to avoid loss of juxtapleural
nodules, this method performs the rolling ball algorithm. Julian Ker [53] has pre-
sented the method of doing segmentation of lungs, which is named as TRACE
method. Due to the possible presence of various disease processes, and the change
of the anatomy with vertical position results in variation of size, shape, texture
of lungs CT image of different patients. Therefore, the boundary between lung
and surrounding tissues can vary from a smooth-edged, sharp-intensity transition
to irregularly jagged edges with a less distinct intensity transition. The TRACE
algorithm implemented with new perception of a non-approximating technique for
edge detection. Shiying et al. [54] have introduced a fully automatic method for
identifying lungs in 3D pulmonary X-Ray CT images. The method follows three
main steps:
• lung region is extracted from CT-Scan image by applying graylevel thresh-
olding,
• by using a dynamic programming it identifies the anterior and posterior
junction, to separate left and right lungs and
• to smooth the irregularities of boundary along the mediastinum nodule, it
implements sequence of morphological operations
Ayman El-Baz et al. [55] have employed a fully automatic Computer-Assisted
Diagnosis (CAD) system for lung cancer screening using chest spiral CT scans.
This paper presents a system for detection of abnormalities, identification or clas-
sification of these abnormalities with respect to specific diagnosis, and provides
the visualization of the results over computer networks. The process of detec-
tion of abnormalities, identification of these abnormalities can achieve by image
analysis system for 3-D reconstruction of the lungs. Riccardo Boscolo et al. [56]
proposed method that uses the novel segmentation technique that combines a
knowledge based segmentation system with a sophisticated active contour model.
This method performs robust segmentation of various anatomic structures. In this
approach the user, need to provide initial contour placement, and the required
parameter optimization automatically determined by the high-level process. Bin-
sheng et al. [57] reported the algorithm, which used the method of selecting the
26
Page 36
3.2 Medical Image Watermarking
threshold value by analyzing the histogram. This algorithm initially separates the
lung parenchyma from the other anatomical structures from the CT images by
using threshold value. By this algorithm structure in CT scan image with higher
densities having some higher density nodules, can grouped into soft tissues and
bones, leading to an incomplete extraction of lung mask. For having complete
hollow free lung mask, morphological closing is applied in this approach. Hossein
B. et al. [45] has introduced the model-based segmentation algorithm. In this
approach instead of using model information to direct the segmentation algorithm
for segmenting an organ of CT scan images, it uses this information to choose a
segment with highest fidelity to the organ.
After completing with the segmentation of ROI, needs to proceed with medical
image watermarking technique to provide security, authentication and privacy of
this medical data. The next section of this paper provides the survey of different
available medical image watermarking approaches.
3.2 Medical Image Watermarking
There has been fair amount of work done in the area of medical image processing.
Numbers of medical image watermarking schemes are reported in this literature
survey, to address the issues of medical information security, and authentication.
Wakatani [58] presented a medical image watermarking, in order not to com-
promise with the diagnosis value, it avoids embedding watermark in the ROI. In
this algorithm watermark to be embed is firstly compressed by progressive coding
algorithm such as Embedded Zero Tree Wavelet (EZW). Embedding process is
done by applying Discrete Wavelet Transform (DWT), for transforming the orig-
inal image using Haar basis. Extraction of watermark is reverse of embedding
process. The major drawback of this algorithm is ease of introducing copy attack
on the non-watermarked area. Yusuk Lim et al. [59] reported a web-based image
authentication system, they used the CT scan images. This technique is mainly
based on the principal of verifying the integrity and authenticity of medical im-
ages. In this approach, the watermark is preprocessed by using 7 most significant
bit-planes except least significant bit (LSB) plane of cover medical image, as a
input to the hash function. This hash function generates binary value of 0 or 1
using secrete key, which is then embedded in LSB bit of cover image to get wa-
termarked image.
27
Page 37
3.2 Medical Image Watermarking
Rodriquez et al. [60] proposed a method in which it searches a suitable pixel
to embed information using the spiral scan that, starts from the centroid of cover
image. Then by obtaining the block with its center at the position of selected
pixel, it checks the value of bit to embed. If bit value is 1, then the embedding
information is obtained by changing the luminance value of the central pixel by
adding the gray-scale level mean of the block with luminance of the block. In
addition, if bit value is 0, then luminance value of the central pixel is changed
by subtracting the luminance value of block from the gray-scale level mean of
the block. While in extraction process, the position of marked pixel is obtained
by spiral scan starting from centroid of the cover image. By checking the lu-
minance value of the central pixel with the gray-scale level mean of the block,
embedded bit is identified. Giakoumaki et al. [20] presented a multiple water-
marking method using wavelet-based scheme. The method provides solution to
the number of medical data management and distribution issues, such as data
confidentiality, archiving and retrieval, and record integrity. In this approach up
to 4 level DWT is performed on medical image. The algorithm embeds multiple
watermarks in different level. A robust watermark containing doctors identifica-
tion code is embedded in 4th level as here capacity is not the matter, only required
is the robustness. In third level decomposition, the index watermark (e.g ICD-10
or ACR diagnostic codes) is embedded. The method embeds caption watermark
holding patients personal information in second decomposition level. Moreover, a
fragile watermark is embedded in forth-level decomposition. Extraction process
is reverse of embedding process. Experimentation is done on ultrasounds medical
images.
Hemin Golpira et al. [61] reported reversible blind watermarking. In this ap-
proach during embedding process, firstly by applying Integer Wavelet Transform
(IDWT) image is decomposed into four subbands. By selecting two points, called
thresholds, according to the capacity required for the watermark data, water-
mark is embedded. To get watermarked image Inverse Integer Wavelet Transform
(IIDWT) is applied. In the extraction process, all of these stages are performed
in reverse order to extract watermark as well as host image.
Nisar Ahmed Memonet et al. [62] presented fragile and robust watermarking
technique for medical images. The method embeds two different watermarks, the
robust watermark and fragile watermark in the medical images. The embedding
28
Page 38
3.2 Medical Image Watermarking
process is start with separation of ROI and RONI from medical images. The
robust watermark containing the electronic patient record (EPR), Doctors identi-
fication code (DIC) and 1st bit-plane of ROI by extracting the LSBs is encrypted
by using pseudo random sequence generated by user defined key. Then this resul-
tant watermark is embedded in high frequency coefficient of IWT in RONI part
of medical data. The proposed method generates fragile watermark by creating
the binary image in tiled fashion and then this fragile watermark is cropped off
by the same size as the ROI. The algorithm embeds this fragile watermark into
spatial domain of ROI part of medical image. The extraction process is reverse of
embedding process.
29
Page 39
Chapter 4
Proposed System for Medical
Image Watermarking preserving
ROI
Our approach focuses on embedding watermark in RONI region of medical image
by preserving ROI. This approach helps in isolating ROI region i.e. not to distort
the critical area of medical image, which will be referred by physician for the
diagnosis. The system diagram for this approach is shown in Figure 4.1. The
system process carried away in three stages:
1. Watermark embedding process
2. Watermark extraction process
3. Watermark authentication process
In first phase of system separating the ROI from the original medical image
provides RONI region for embedding watermark. This step isolates ROI from em-
bedding process. In this phase multiple watermarks are embedded into the RONI
area of medical image. Embedding multiple watermarks ensure high security of
medical image as it carries high payload and it will be more complex to break the
system. Here fragile watermarking system is used to get the benefit of identifying
whether a medical image is tampered or not?
After the completion of embedding process the separated ROI is combined
with the produced watermarked medical image. The resultant watermarked med-
ical image is then sent to the receiver.
30
Page 40
4.1 Separating ROI from medical image
Figure 4.1: Medical Image Watermarking Approach Preserving ROI
In watermark extraction phase, first step is separating the ROI from the wa-
termarked medical image. The remaining watermark extraction process is exact
reverse of embedding process, where the embedded watermark will be extracted
from the watermarked medical image. The watermark authentication is achieved
by comparing the extracted watermark with the original watermark. This process
helps in identifying if any tampering or manipulation to the watermarked medical
image over the public network.
4.1 Separating ROI from medical image
As discussed earlier for separating ROI Segmentation method is used. However
segmentation is semi-automatic procedure and it needs to define a seed point in
an image. Therefore, the algorithm, which gives perfect result for one type of
application, may not even work for another.
In proposed system for separating ROI the Graphical User Interface (GUI) is
implemented, so that it will work for all kinds for medical image (such as CT
scan, MRI, X-Ray, Ultrasound, etc.). The interface for the implemented GUI
based approach is shown in Figure 4.2. In this method user has an option to
select the part of medical image (square in size) which has critical information
and used for the reference of physician. This GUI based system returns the Xmin,
31
Page 41
4.1 Separating ROI from medical image
Figure 4.2: Interface for GUI based approach
Xmax, Ymin, Ymax pixel values of selected ROI region and image of selected
ROI. This resulted ROI image can be saved, so that it can be combined with the
resultant watermarked image. The dashed square in Figure 4.2 is the user selected
ROI of medical image, the region that can be selected by mouse click function.
The respective pixel values (Xmin, Xmax, Ymin, Ymax) are shown at top of the
window panel.
Steps in ROI separation approach
• Mouse click function: For selecting the ROI, mouse clicking function is used.
• Done button: To get the output after selection process, done button is im-
plemented.
• Storing handles: For safe storing the pixels values of selected ROI (Xmin,
Xmax, Ymin and Ymax) and image of selected ROI, the storing handles are
use.
• Start button: It is implemented to clear the stored handles to start again
the process of selecting ROI.
• Zooming option: It is provided for zooming the image, so that the image
will be clear to select the ROI.
32
Page 42
4.2 Medical Image Watermarking System
4.2 Medical Image Watermarking System
For the implementation of Medical Image Watermarking, we referred the algo-
rithm proposed by Giakoumaki et al. [20]. The algorithm provides solution to
the number of medical data management and distribution issues, such as data
confidentiality, archiving and retrieval, and record integrity. The medical water-
marking system embeds the multiple watermarks. The watermarks used to embed
are in the text form. In this approach medical image is decompose with 4-level
lifting based DWT transform. The lifting based DWT is a better method to ob-
tain the wavelet transform. For the development of second generation wavelet the
lifting based DWT approach is proposed. Advantage of second generation wavelet
over first generation wavelet is that, it does not use the translation and dilation
of the same wavelet prototype in different levels. Using the Euclidean algorithm
any classical wavelet filter bank can be decomposed into lifting steps. The lifting
based DWT consists of three stages i.e. split, predict and update. In split stage
the input signal x[n] get divided into two subsets i.e. even set s[n] and odd set o[n].
This process is known as lazy wavelet transform. The predict step use the linear
combination of elements in one subset to guess the values of the other subset with
assumption that the subsets produced in the split stage are correlated with each
other. The predicted values would be close to the original values if the correlation
between both the subset is high. Generally the linear combination of the even
subset elements are used to predict odd subset values. The predict step is defined
as:
d[n] = o[n]−∑k
p[k]s[n− k] (4.1)
Where d[n] is the difference between the actual values and the predicted values,
P[k] is prediction coefficient. Although there are chances of loss of properties of
signal such as mean value in the predict step.
s1[n] = s[n]−∑k
u[k]d[n− k] (4.2)
The predict step causes the loss of some basic properties of the signal like mean
value, which needs to be preserved. The update step lifts the even sequence values
using the linear combination of the predicted odd sequence values so that the basic
properties of the original sequence is preserved [5]. The even sequence values s1
obtained as the result of equation 4.2 is equivalent to the sub-sampled low pass
version of the original sequence.
33
Page 43
4.3 Method
4.2.1 Integer to Integer transform
It was observed that usually when wavelet transforms is performed on integer se-
quence it gives floating point coefficients. As per Calderbank [6] wavelet transform
which will map integers to integers can be build with the help of lifting structure.
This can be achieved by rounding off or updating the filter in each lifting step
before its addition or subtraction. The invert of the lifting steps can be produced
by following the exact reverse operation and flipping the signs.
4.3 Method
In recent days the wavelet analysis got a good recognition in research and devel-
opment area due to its characteristic of providing time and frequency information
simultaneously. As per research the retina of the eye splits an image in to several
frequency channels i.e. approximately one octave. In multi resolution decomposi-
tion, the image is divided into bands of equal bandwidth on a logarithmic scale.
There is lot of similarity between the signal processing of the human visual system
(HVS) and scaling decomposition of the wavelet transform, which can be achieved
by watermark embedding to the masking property or quantization method [7].
4.3.1 Description
The watermarks used in this approach:
1. Doctor’s identification code
2. Indexed watermark
3. Patient’s reference identification code
4. Patient’s diagnosis information
5. Patient’s treatment information
The listed watermarks used in this proposed watermarking scheme helps in ad-
dressing different issues and concerns in healthcare management system, Such as
confidentiality of medical data, recovering original image without any distortion,
data integrity, authentication and efficient data management.
34
Page 44
4.3 Method
Confidentiality of medical data is achieved by embedding watermark using
Integer to Integer Discrete Wavelet Transform (IDWT), which confirms the im-
perceptibility property. This property ensures the embedded watermark will be
invisible to the normal human eye and the watermark can be extracted by the
one who knows the embedding and extraction algorithm applied in this system.
By applying Inverse IDWT at the receiver end original image can be recovered
without any distortion. Also the distortion to the ROI has already been avoided
by separating the ROI before embedding the watermark in to the medical image.
Medical data integrity is achieved by using fragile watermarking system, so any
manipulation on medical image data leads in distortion of embedded watermark.
For the authentication purpose the included watermarks such as doctor’s identifi-
cation code, patient’s identification code will ensures the entitled users can access
or modify the medical data. To provide efficient data management in this system
the indexed watermark is embedded which helps in retrieving the image for the
future reference if needed using database query.
The watermarks are inserted in different decomposition levels and sub-bands
depending on their type. They can be independently embedded and retrieved
without any intervention among them. By integrating this idea in to different
medical acquisition systems like Ultrasound, CT and MRI etc. This system can
be applied in different applications such as e-diagnosis or medical image sharing
through picture archiving and communication.
Selection of embedding coefficient
The Figure 4.3 illustrates the subband structure of a 4-level harr wavelet decompo-
sition of an original medical image, which is obtained after removing the ROI from
the host medical image. This decomposed image comprise of a coarse scale image
approximation at the highest decomposition level i.e. at 4th level, and it also
contains the twelve detail images corresponding to the horizontal (HL), vertical
(LH), and diagonal (HH) details at each of the 4-level.
The watermark holding the doctor’s identification is the most important for the
identification purpose and are of limited in length hence capacity is not very im-
portant. By considering this two points this doctor’s identification code containing
watermark is embedded in the fourth level, because more the decomposition level
more the robust watermark. On the other hand index watermark and patient’s
identification code requires more space than the doctor’s identification code since
they pass on many bits of additional information. The index watermark holds the
35
Page 45
4.3 Method
decomp.png
Figure 4.3: Sub-band structure a of 4-level wavelet transform
information, which is used to retrieve the medical image. So the required capac-
ity for this watermark lies between the degrees of capacity intended for patient’s
identification code, patient’s diagnosis information, and treatment information.
Focusing on the above fact indexed watermark is embedded into third decompo-
sition level. The watermark contain patient’s identification is embedded into 2nd
decomposition level as it requires less space than both diagnosis information water-
mark and treatment watermark. The patient’s diagnosis information watermark
and patient’s treatment information watermark both requires the more capacity
so these fragile watermarks are embedded into 1st level. If any modification or
tampering to the embedded image occurs then the extracted watermark will be
totally different than embedded one.
In general horizontal and vertical sub-bands are used to embed the watermark,
as they have more or less same behavior in contrast to diagonal one. By embed-
ding the watermark into horizontal or vertical details coefficients results in less
distortion of image. Especially for the ultrasound images the energy of horizontal
details is more than compare to the vertical and diagonal details. This is due to
36
Page 46
4.3 Method
Table 4.1: Energy of Approximation and Detail Images of Four LevelDWT
Sub-bands Level 1 Level 2 Level 3 Level 4Approximation 41.4514
Horizontal 5.4363 8.8305 14.8038 16.9878Vertical 4.5678 7.2816 9.0938 10.4323Diagonal 3.3392 5.6763 9.3863 12.5729
the elongation of ultrasound image mark spots in the horizontal direction [57].
For all other medical images the energy of horizontal and vertical details are ap-
proximately same. Table 4.1 shows the energy of the approximation and detail
images for a 4-level haar wavelet decomposition of an ultrasound test image shown
in Figure 4.4.
Figure 4.4: Ultrasound Image
From the table 4.1, it is clear that at the higher decomposition level corre-
sponding to the low frequency coefficients have the more energy than the energy
bat the lower decomposition level. Moreover the energy of horizontal details sub-
band is more than the energy of vertical and diagonal details sub-band. Hence
the watermarks are embedded into horizontal sub-bands in this system. As the
approximation sub-bands LL4 has most energy of the medical image and has the
huge amount of impact on the quality of medical image, it is not used for embed-
ding purpose to retain imperceptibility.
The energy of the approximation and detail images obtained by four-level DWT
can be calculated as:
ek =1
NkMk
∑i
∑j
|Ck (i, j)| (4.3)
Where k is the approximation and detail images at each of the decomposition
levels, Ck are the coefficients of the sub-band images, Mk and Nk are their corre-
sponding dimensions.
37
Page 47
4.3 Method
4.3.2 Algorithm
In this algorithm the multiple watermarks embedding technique is used. Where,
depending on the quantization of selected coefficients the multiple watermarks
embedding procedure is used. This prevents any modification to the watermark
bits by granting integer changes in spatial domain of medical image. This can
be achieved by applying 4-levl haar wavelet transform to decompose the host
medical image. Moreover it gives the output as coefficients, which are in the
form of dyadic rational numbers. These coefficients denominators are in powers
of 2. The multiple of 2l (l is decomposition level) number adding or subtracting
to the produced coefficient value, assures that the inverse DWT provide integer
pixel values. Wavelet transform generally provides the coefficients which are real
numbers. By applying the quantization function it assigns the binary number to
every coefficient. This quantization function is defined as
Q(f) = 0, if
⌊(f − s
∆
)⌋is even (4.4)
Q(f) = 1, if
⌊(f − s
∆
)⌋is odd (4.5)
Where s is a user-defined offset for increased security, f is frequency coefficient
produced by haar wavelet transform and , the quantization parameter, is a positive
real number. Moreover ∆ is defined as ∆ =2l. The quantization procedure is
shown in Figure 4.5.
Figure 4.5: Quantization Procedure
As explain earlier, addition or subtraction of a multiple of 2l value to the haar
wavelet coefficient results in integer pixel values, after applying inverse DWT. Dur-
ing the embedding process the algorithm add or subtract an appropriate constant
to the haar coefficient chosen for watermark casting.
The algorithm for embedding multiple watermarks is explained below:
Step 1: Separate the ROI region from the host medical data using GUI based
38
Page 48
4.3 Method
mouse clicking approach. Which results in image of RONI region, name it as orig-
inal medical image.
Step 2: Save the removed ROI from medical image.
Step 3: The multiple watermarks to be embed into a original image is generated
by reading the patient’s information file from text document, and converting it
into binary.
Step 4: Apply the 4-level Haar-lifting wavelet transform to original medical im-
age, to obtained a gross image approximation at the lowest resolution level and a
sequence of detail images corresponding to the horizontal, vertical, and diagonal
details at each of the four decomposition levels.
Step 5: On each decomposition level the watermark bit wi is embedded into
the key determined coefficient f, which is obtained by applying wavelet transform
according to the following condition:
1. If Q(f) = wi, the coefficient is not modified
2. Otherwise, the coefficient is modified so that Q(f) = wi, using the following
equation:
f = f + ∆, if f ≤ 0 (4.6)
f = f −∆, if f > 0 (4.7)
Step 6: The pre watermarked image is produced by performing the corresponding
four level inverse wavelet transform.
Step 7: The resultant watermarked image is obtained by combining the saved
ROI with the pre watermarked image.
The watermark extraction process is similar to that of embedding one except
that at the receiving end extractor should have the knowledge of location of the
embedded watermark. This can achieve by the key-based embedding and detec-
tion. With this type of method access to the watermark by unauthorized users is
prevented. The algorithm for extraction process to recover the host medical image
is explained below.
Step 1: Remove the ROI region from the received watermarked image with the
help of Xmax, Xmin, Ymin and Ymax parameter provided with watermarked im-
age.
Step 2: Apply the 4-level lifting-haar wavelet transform to the image which is
created from step 1, which results in a image approximation at level four and
sequence of images corresponding to the horizontal, vertical, and diagonal details
39
Page 49
4.3 Method
at each of the four decomposition levels.
Step 3: Identify the location of watermark by key-based detection.
Step 4: Extract the watermarks by applying quantization function defined in
equation 4.4 and 4.5, which recovers the original coefficient. Convert the ex-
tracted binary watermark to text watermark.
Step 5: The pre output image is obtained by applying inverse 4-level haar wavelet
transform.
Step 6: combine the separated ROI region to the pre output image to get the
original host medical image.
40
Page 50
Chapter 5
Experiments and Results
The proposed system has been applied against different type of medical image
such as, CT scan, MRI, X-Ray and Ultrasound. We have tested the system over
different size of medical images like 320 X 256, 384 X 384, and 512 X 512.
The applied watermark was consists of in
1. Doctor’s identity: G123468
2. Indexing for database: 321-123.1
3. Patient’s identification: sonika c rathi.190.85.04567851
4. Diagnosis Information: light.sugar healthy extra.spicy no.fats 12189.75.1
5. Treatment applied to the patient: painkiller.hgkkfgjklfd abcdefmglkh bkjdhflkds.yeio
The results after applying the system against CT scan, MRI, X-Ray and Ultra-
sound are shown below:
5.1 The experiments and results of the system
without attacks
5.1.1 CT Scan Images
The embedded and extracted watermark values are shown in Figure 5.4, with
PSNR value (p) and MSE (d).
The system is applied on different CT scan images considering their image
size and noted corresponding results, which are shown in table 5.1. The table
shows the PSNR value for ROI extracterd from host image and ROI extracted
41
Page 51
5.1 The experiments and results of the system without attacks
Figure 5.1: Segmenatted ROI of host image
Figure 5.2: (a) The original host CT scan image, (b)Roi removed im-age,(c)Emebedded image without ROI, (d)Final embedded image with ROI
Figure 5.3: Recovered original image
from watermarked image. As the corner pixel values of ROI image is changed the
PSNR is not ∞ but there correlation is approximately 1. So, the selected ROI
should be large enough to not compromise with the diagnosis value. The table
also provides the PSNR value for embedded image and original image and there
respective mean square difference.
42
Page 52
5.1 The experiments and results of the system without attacks
Figure 5.4: Embedded and extracted watermark values without any attacks
Table 5.1: Results of CT scan imagesSize MSE PSNR PSNR extracted ROI Correlation between the ROI
320 X 256 6.81 39.83 38.69 0.9963384 X 384 4.00 42.23 43.20 0.9994512 X 512 2.11 44.71 45.12 0.9998
5.1.2 MRI Images
The embedded and extracted watermark on the MRI image shown in Figure 5.5
(a), are given in Figure 5.6.
The results after applying the system on different size of MRI images are shown
in table 5.2.
Table 5.2: Results of MRI imagesSize MSE PSNR PSNR extracted ROI Correlation between the ROI
320 X 256 6.86 39.40 37.24 0.9962384 X 384 3.84 42.28 41.20 0.9993512 X 512 2.22 44.70 44.12 0.9998
43
Page 53
5.1 The experiments and results of the system without attacks
Figure 5.5: (a) The original host MRI image, (b)ROI image of host image,(c)Roiremoved image (c)Emebedded image without ROI, (d)Final embedded image withROI
, (e)Recovered image
Figure 5.6: Embedded and extracted watermark values without any attacks forMRI image
5.1.3 X-Ray Images
The embedded and extracted watermark on the X-Ray image shown in Figure 5.7
(a), are given in Figure 5.8.
44
Page 54
5.1 The experiments and results of the system without attacks
Figure 5.7: (a) The original host X-Ray image, (b)ROI image of host image,(c)Roiremoved image (c)Emebedded image without ROI, (d)Final embedded image withROI
, (e)Recovered image
Figure 5.8: Embedded and extracted watermark values without any attacks forX-Ray image
The results after applying the system against different size of X-Ray images
are shown in table 5.3.
45
Page 55
5.2 Embedded and extracted watermark with attacks
Table 5.3: Results of X-Ray imagesSize MSE PSNR
320 X 256 6.93 39.60384 X 384 3.92 42.18512 X 512 2.16 44.63
5.1.4 Ultrasound Images
Figure 5.9: (a) The original host Ultrasound image, (b)ROI image of host im-age,(c)Roi removed image (c)Emebedded image without ROI, (d)Final embeddedimage with ROI
, (e)Recovered image
The embedded and extracted watermark on the Ultrasound image shown in
Figure 5.9 (a), are given in Figure 5.10.
The results after applying the system against different size of Ultrasound im-
ages are shown in table 5.4.
5.2 Embedded and extracted watermark with at-
tacks
The embedded and extracted watermark after applying different attacks on the
watermarked images are shown in this section. The attacks applied on the water-
46
Page 56
5.2 Embedded and extracted watermark with attacks
Figure 5.10: Embedded and extracted watermark values without any attacks forUltrasound image
Table 5.4: Results of Ultrasound imagesSize MSE PSNR
320 X 256 6.93 39.60384 X 384 3.92 42.18512 X 512 2.16 44.63
marked medical images are:
• Slat and pepper noise attack
• Cropping attack
• Histogram equalization
• Sharpning attack
• Sampling attack
• JPEG compression attack
The following figures shows the attacked watermark medical image and em-
bedded, extracted watermark of CT scan, MRI, X-Ray and Ultrasound images:
47
Page 57
5.2 Embedded and extracted watermark with attacks
Figure 5.11: (a) The original watermarked CT scan image, (b)The image aftersharpning attack with 0.02 factor
The extracted watermark from the attacked CT scan image is shown in Figure
5.12. The sharpning attack with 0.02 factor is applied on watermarked CT scan
image, shown in Figure 5.11 (a).
Figure 5.12: Embedded and extracted watermark values with sharpning attck(0.02 factor) image
The Figure 5.13 shows the original watermarked MRI image and the histogram
attacked watermarked image. The embedded and extracted watermark values
48
Page 58
5.2 Embedded and extracted watermark with attacks
after histogram attack are given in Figure 5.14.
Figure 5.13: (a) The original watermarked MRI image, (b)The image after His-togram attack
Figure 5.14: Embedded and extracted watermark values after histogram attackon image
As, the algorithm implemented is fragile system, after even the 10% of jpeg
compression to the X-Ray image, the extracted watermark from attacked image is
totally distorted. This distortion of extracted watermark is shown in Figure 5.16,
the original watermarked image and attacked image is shown in Figure 5.15.
49
Page 59
5.2 Embedded and extracted watermark with attacks
Figure 5.15: (a) The original watermarked X-Ray image, (b)The image after 10%JEPG compression attack
Figure 5.16: Embedded and extracted watermark values after JEPG Compressionattack on X-Ray image
The Figure 5.17 shows the original watermarked image and attacked water-
marked image of ultrasound image. Here the Down and Up sampling attack is
applied on watermarked ultrasound image. The attack is applied with 1 factor of
50
Page 60
5.2 Embedded and extracted watermark with attacks
down sampling and 1 factor of up sampling on the watermarked image. The Fig-
ure 5.17 clearly shows that both the images are look like same, that is by normal
human eye the difference between the two image is not visible. However the ex-
tracted watermark from the attacked image is totally different than the embedded
one. The embedded and extracted watermark values are shown in Figure 5.18
Figure 5.17: (a) The original watermarked Ultrasound image, (b)The image afterup and down sampling attack
51
Page 61
5.2 Embedded and extracted watermark with attacks
Figure 5.18: Embedded and extracted watermark values after down and up sam-pling attack on Ultrasound image
52
Page 62
Chapter 6
Conclusion and Future Work
6.1 Conclusion
There exist various medical image watermarking algorithms which provide the
confidentiality of medical data, recovering original image without any distortion,
data integrity, authentication and efficient data management. Also the different
segmentation algorithms are in place, which vary for the types of medical images
such as MRI, CT scan, X-ray and Ultrasounds etc.
Here the proposed system used an algorithm to separate ROI from the host
medical image that will be applicable for all types of medical images. Separated
ROI can be stored with xmin, xmax, ymin, and ymax value so that at the end of
embedding process before transmitting watermarked image, the segmented ROI
can be attached with watermarked image. And the ROI region which is considered
as a critical data and used as a reference by the physician for the treatment will
be safe.
6.2 Future Work
Proposed system uses DWT approach for embedding the watermark, instead of
DWT use of Complex Wavelet Transform (CWT) will make the system more ro-
bust and secure.
The current proposed system can further be extended to provide more secured
system. This can be done by encrypting the watermark using secret key, before
embedding it in to medical images. Having the automated tool for separating the
53
Page 63
6.2 Future Work
ROI from medical image will provide faster system and more accurate system,
which will be easier for end user.
The watermark before embedding can be compressed and then embedded. This
will lead to more secured system. Also, it will take more effort to break the system.
54
Page 64
Bibliography
[1] From Wikipedia http://en.wikipedia.org/wiki/Digital_watermarking.
[2] I. J. Cox, M. L. Miller, J. A. Bloom, ”Digital Image Watermarking”, Morgan
Kaufman, Publishers, USA, 2004.
[3] H. C. Huang, H. M. Hang, J. S. Pan, ”An Introduction to Watermarking
Techniques”, Series on Innovative Intelligence, H. C. Huang, H. M. Hang, L.
C. Jain (Eds), World Scientific, Vol. 7, pp. 3-39, 2004.
[4] J. T. Brassil, S. Low, N. F. Maxemchuck, ”Copyright Protection for Electronic
Distri-bution of Text Documents”, Proceedings of IEEE, Vol. 87, No. 7, pp.
1181-1196, July 1999.
[5] I. J. Cox, J. Kilian, F. T. Leighton, T. Shamoon, ”Secure Spread Spectrum
Water-marking for Multimedia”, IEEE Transactions on Image Processing, Vol.
6, No. 12, pp. 1673-1686, December 1997.
[6] M. D. Swanson, B. Zhu, A. H. Tewfik, L. Boney, ”Robust Audio Watermarking
using Perceptual Masking”, Signal Processing, Vol. 66, No. 3, pp. 337-355, May
1998.
[7] F. Hartung, B. Girod, ”Watermarking of Uncompressed and Compressed
Video”, Signal Processing, Vol. 66, No. 3, pp. 283-301, May 1998.
[8] O. Benedens, ”Geometry-based Watermarking of 3D Models”, IEEE Computer
Graph-ics and Applications, Vol. 19, No. 1, pp.46-55. 1999.
[9] M. G. Wagner, ”Robust Watermarking of Polygonal Meshes”, Proceedings of
Geometric Modeling and Processing 2000: Theory and Applications, pp. 201-
208, 10-12 April, 2000.
[10] P.-C. Su, H. Wang, C.-C. J. Kuo, ”Digital Image Watermarking in Region
of Interest”, IS&T’s Image Processing, Image Quality, Image Capture (PICS)
Conference, Georgia, April 1999.
55
Page 65
BIBLIOGRAPHY
[11] C.-S Woo, ”Digital Image Watermarking Methods for Copyright Protection
and Authentication”, PhD Thesis, Queensland University of Technology, Aus-
tralia, March 2007.
[12] D. Zheng, Y. Liu, J. Zhao, A. Saddik, ”A Survey of RST Invariant Image
Water-marking Algorithms”, ACM Computing Surveys Vol. 39, No. 2, Article
5, 91 pages, 2007.
[13] M. Barni, F. Bartolini, ”Watermarking Systems Engineering”, Signal and
Commmu-nication Series, Marcel Dekker Inc. USA, 2004.
[14] Munesh Chandra”, Shikha Pandel, Rama Chaudharl Digital Watermark-
ing Technique for Protecting Digital Images”, R.K.G. I.T, Ghaziabad
a,bGhaziabad, India.
[15] X. Wu, Z.-H. Guan, Z. Wu, ”A Chaos Based Robust Spatial Domain Water-
marking Algorithm”, Spring Verlog, LNCS, 4492, pp. 113-119, 2007.
[16] F. Sebe, T. Domingo-Ferrer, J. Herrera, ”Spatial Domain Image Wateram-
rking Robust against Compression, Filtering, Cropping and Scaling”, Springer
Verlog, LNCS, 1975, pp. 44-53, 2000.
[17] Shikha Tripathi, ”Novel DCT and DWT based Watermarking Techniques for
Digital Images”, R.C. Jain Birla Institute of Technology & Science, Pilani
Rajasthan, India. V. Gayatri HP LABS Bangalore, India.
[18] I. J. Cox, J. Kilian, F. T. Leighton, T. Shamoon, ”Secure Spread Spectrum
Watermarking for Images, Audio and Video”, IEEE International Conference
on Image Processing, pp. 243-246, 1996.
[19] C.-T. Hsu, J.-L. Wu, ”Hidden Digital Wateramrks in Images”, IEEE Trans-
actions on Image Processing, Vol. 8, pp. 58-68, 1999.
[20] Giakoumaki, Sotiris Pavlopoulos, and Dimitris Koutsouris, ”Multiple Image
Watermarking Applied to Health Information Management”, IEEE Trans. on
information technology in biomedicine, vol. 10, no. 4, Oct. 2006.
[21] Imen Fourati Kallel, Mohamed Kallel, Mohamed Salim BOUHLEL, ”A Secure
fragile Watermarking Algorithm for medical Image Authentication in the DCT
Domain”, IEEE 2006.
56
Page 66
BIBLIOGRAPHY
[22] M.S.Bouhlel, ”Conception d’une banque d’images medicales sur INTER-
NET”, 3eme Rencontres Institutionnelles: Rhones Alpes/ Tunisie (RI-
RAT’02). Tozeur, Tunisie, 21-22, 2002.
[23] G. Coatrieux, H. Maitre, B. Sankur, Y. Rolland, R. Collarec, ”Relevance
of Water-marking in Medical Imaging”, Proceedings of IEEE-EMBS Interna-
tional Conference on Information Technology Applications in Bio Medicine,
pp. 250-2555, 9-10 November, 2002.
[24] H.K. Wu, R.-F Chang, C.-J. Chen, C.-L. Wang, T.H. Kuo, W. K. Moon, D.-R.
Chen, ”Tamper Detection and Recovery for Medical Images Using Nearlossless
Information Hiding Technique”, Journal of Digital Imaging, Vol. 21, No. 1, pp.
59-76, March 2008.
[25] V. Fotopoulos, M. L. Stavrinou, A. N. Skodras, ”Medical Image Authentica-
tion and Self-Correction through an Adaptive Reversible Watermarking Tech-
nique”, Proceedings of 8th IEEE International Conference on Bio-Informatics
and Bio-Engineering (BIBE-2008), pp. 1-5, October 2008.
[26] Preeti Aggarwal, Renu Vig, Sonali Bhadoria, and C.G.Dethe , ”Role of Seg-
mentation in Medical Imaging: A Comparative Study”, International Journal
of Computer Applications (0975 8887), Volume 29 No.1, September 2011.
[27] Pradeep Singh, Sukhwinder Singh, Gurjinder Kaur, ”A Study of Gaps in CB-
MIR using Different Methods and Prospective”, Proceedings of world academy
of science, engineering and technology, volume 36 , ISSN 2070-3740, pp. 492-
496, 2008.
[28] Zhen Ma, Joao Manuel, R. S. Tavares, R. M. Natal Jorge, ”A review on the
current segmentation algorithms for medical images”, 1st International Con-
ference on Imaging Theory and Applications (IMAGAPP), Lisboa, Portugal,
INSTICC Press, pp. 135-140, 2009.
[29] Nisar Ahmed Memon, Anwar Majid Mirza, and S.A.M. Gilani, ”Segmentation
of Lungs from CT Scan Images for Early Diagnosis of Lung Cancer”, World
Academy of Science, Engineering and Technology 20, 2006.
[30] M. M. Khalighi, H.S.Zadeh, C. Lucas, ”Unsupervised MRI segmentation with
spatial connectivity”, 23-28, San Diego, CA. ”in press”, Feb 2002.
57
Page 67
BIBLIOGRAPHY
[31] L. Jiang, W. Yang, ”A modified fuzzy c-means algorithm for segmentation of
MR Images”, Proc. VIIth Digital Image Computing: Techniques and Appli-
cations. , 10-12, Sydney, ”in press” Dec 2003.
[32] M.Y. Siyal, L. Yu, ”An intelligent modified fuzzy c-means based algorithm for
bias estimation and segmentation of brain MRI Pattern Recognition Letters”,
pp. 20522062, 2005
[33] Moslem Taghizadeh, Mahboobeh Hajipoor, ”A Hybrid Algorithm for Segmen-
tation of MRI Images Based on Edge Detection”, 2011 International Confer-
ence of Soft Computing and Pattern Recognition (SoCPaR), 2011.
[34] M.Sezgin, ”Survey over image thresholding techniques and quantitative per-
formance evaluation”, Journal of Electronic Imaging 13(1), pp.146-165, 2004.
[35] Yavuz, Z., ”Comparing 2D matched filter response and Gabor filter methods
for vessel segmentation in retinal images”, IEEE Trans. Electrical, Electronics
and Computer Engineering (ELECO), vol.8, no.6,pp.648-652, 2010.
[36] Dong-Chul Park, ”Intuitive Fuzzy C-Means Algorithm for MRI Segmenta-
tion”, 978-1-4244-5943-8/10/, IEEE, 2010.
[37] J.Bezdek, ”Pattern recognition with fuzzy objective function algorithms”,
Plenum, New York, 1981.
[38] P. Wang and H. Wang, ”A Modied FCM Algorithm for MRI B rain Image
Segmentation”, Proc. Fut. Biomed. Info. Eng., 26-29, 2008.
[39] R. Krishnapuram and J. Keller, ”A possibilistic approach to clustering”, IEEE
Trans. Fuzzy Syst., 1(2), 98-110, 1993.
[40] N. Pal, K. Pal, and J. Bezdek, ”A Possibilistic Fuzzy c-Means Clustering
Algorithm”, IEEE Trans. Fuzzy Sys., 13(4), pp. 517-530, 2005.
[41] Talib Hussein R., ”Automatic Extracted Object Technique for Contrast En-
hancement Medical Images”, IJCCCE, VOL.9, NO.1, 2009.
[42] Civicioglu P., ”CCII based analog circuit for the edge detection of MRI im-
ages”, IEEE Trans. Micro-NanoMechatronics and Human Science, vol.1, no.6,
pp.341-344, 2003.
[43] Sachin G Bagul, ”Comparison of SUSAN and Sobel Edge Detection in MRI
Images for Feature Extraction”, IJCA Journal, VOL.1, NO.1 , USA, 2011.
58
Page 68
BIBLIOGRAPHY
[44] J. F. Canny, ”A computational approach to edge detection”, IEEE Trans.
Pattern Analysis and Machine Intelligence, vol.8, no.6, pp.679-698 1986.
[45] Hossein Badakhshannoory and Parvaneh Saeedi, ”A Model-Based Valida-
tion Scheme for Organ Segmentation in CT Scan Volumes”, IEEE Trans. on
biomedical information, vol. 58, no. 9, September 2011.
[46] R. Susomboon, D. Raicu, and J. Furst, ”A hybrid approach for liver segmen-
tation”, in Proc. 3-D Segment. Clin.-MICCAI Grand Challenge 2007.
[47] K. Seo, L. C. Ludeman, S. Park, and J. Park, ”Efficient liver segmentation
based on the spine”, Adv. Inf. Syst., vol. 3261, pp. 400409, 2005.
[48] A. H. Forouzan, R. A. Zoroo, M. Hori, and Y. Sato, ”Liver segmentation
by intensity analysis and anatomical information in multislice CT images”, in
Proc. Liver Segment. Intensity Anal Anatomical Inf. Multi-Slice CT Images,
vol. 4, pp. 287297, 2009.
[49] S. Pan and B. M. Dawant, ”Automatic 3D segmentation of the liver from ab-
dominal CT images: A level-set approach”, Proc. SPIE, vol. 4322, pp. 128138,
2001.
[50] D. T. Lin, C. C. Lei, and S. W. Hung, ”Computer-aided kidney segmentation
on abdominal CT images”, IEEE Trans. Inf. Technol. Biomed.,vol. 10, no. 1,
pp. 5965, Jan 2006.
[51] H. Badakhshannoory and P. Saeedi, ”Liver segmentation based on de-formable
registration and multilayer segmentation”, in Proc. IEEE Int. Conf. Image
Process., pp. 25492552, 2010.
[52] Samuel G. Armato III, Maryellen L. Giger and Catherine J. Moran, (1999)
”Computerized Detection of Pulmonary Nodules on CT Scans”, RadioGraph-
ics, vol. 19, pp. 1303-1311, 1999.
[53] Julian Kerr, ”The TRACE method for Segmentation of Lungs from Chest
CT images by Deterministic Edge Linking”, University of New South Wales,
Department of Artificial Intelligence, Australia, May 2000.
[54] Shiying Hu, Eric A.Huffman, and Joseph M. Reinhardt, ”Automatic Lung
Segementation for Accurate Quantitiation of Volumetric X-Ray CT images”,
IEEE Transactions on Medical Imaging, vol. 20, No. 6, June 2001.
59
Page 69
BIBLIOGRAPHY
[55] Ayman El-Baz, Aly A. Farag, Robert Falk, and Renato La Rocc, ”Detection,
Visualization, and Identification of Lung Abnormalities in Chest Spiral CT
Scans: Phase 1”, International Conference on Biomedical Engineering, Cairo,
Egypt, Jan 2002.
[56] Riccardo Boscolo, Mathew S. Brown, Michael F. McNitt-Gray, ”Medical Im-
age Segmentation with Knowledge-guided Robust Active Contours”, Radio-
graphics, vol. 22, pp. 437-448, 2002.
[57] Binsheng Zhao, Gordon Gamsu, Michelle S. Ginsberg, ”Automatic detection
of small lung nodules on CT utilizing a local density maximum algorithm”,
Journal of Applied Clinical Medical Physics, vol. 4, No. 3, 2003.
[58] A. Wakatani, ”Digital Watermarking for ROI Medical Images by Using Com-
pressed Signature Image”, Proceedings of the 35th International Conference
on System Sciences, Jan 2002.
[59] Yusuk Lim, Changsheng Xu, and David Dagan Feng, ”Web based Image Au-
thentication Using Invisible Fragile Watermark”, Pan-Sydney Area Workshop
on Visual Information Processing (VIP2001), Sydney, Australia.
[60] C.R. Rodriguez, F. Uribe Claudia, T. Blas Gershom De J, ”Data Hiding
Scheme for Medical Images”, IEEE 17th International Conference on Elec-
tronics, communications and computers, 2007.
[61] Hemin Golpira and Habibollah Danyali, ”Reversible Blind Watermarking for
Medical Images Based on Wavelet Histogram Shifting”, IEEE, 2009.
[62] Nisar Ahmed Memon, S.A.M. Gilani, and Shams Qayoom, ”Multiple Water-
marking of Medical Images for Content Authentication and Recovery”, IEEE,
2009.
60